Aug. 18, 2022
Read time: 3 minutes and 2 seconds.
tags:So, everyone’s bowing down to AI, and I’m usually first in line to play with new tech. But here’s the deal: AI art generators, at this stage, are kind of a mess. Unless you’re a prompt whisperer, you’re in for a world of frustration. I can definitely speak from experience as I’ve kicked the tires on my fair share of technology, especially as it’s released.
This is one of those experiences where you have a certain expectation of something you want to happen, but then it turns sideways. Like surprising your crush on a given day so that she can crush you by saying, “hey, I hope I never see you again”. The big challenge with out of the box AI like Dall-E or Stable diffusion is the upstart challenge (if you’re not familiar with technology). But if you’re familiar with technology, you’ll definitely run into big time hurdles with standard AIs is the pre-moderation. They add this fatal cushion that can adjust and adapt the image that you’re trying to draw in ways you probably never wanted it to.
To some degree, I can understand why it happens. You have different skill sets and the people who are trying to create images or media. So, the guy who knows what he’s doing is bucketed with those who don’t. It might sound helpful in theory, but it hurts more than it helps when you’re looking to generate consistent outputs for a scene, a character or virtually anything else.
For example, you type “dog,” and the AI adds all sorts of fluff to guide it. Great for basic stuff, terrible if you want, say, a specific breed in a particular pose.
If you’re trying to create a consistent character in a story for example, you’ll deal with this hair pulling structure trying to make the same character persistent. As much as some of the use cases that these companies promote and sell are stories and drawing for kids, the challenge becomes an evolving character and one’s lack of photoshop skills. I mean, I am that MS paint guy. I wouldn’t ever trust my editing skills… under any circumstance really. In one example, I tried to make a sequence of images with the same person appearing in multiple images and I couldn’t really craft a storyline…very different people and then we had variable images on top of that too. It was like changing the main character in a TV show, you’re going to notice and ask questions… and spend a ton of money in credits trying to perfect it.
It is cool that you can make hyper-realistic images. However, it is nowhere near replacing humans or even looking human-like even with the best of prompts.
You can get amazing results, but it takes hours of tweaking prompts. Plus, what actually worked and helped? Spending a bunch of time on crafting a list of parameters that can illustrate an image on your behalf as cohesively as possible. Down to the lighting and everything. Like, just not something that could give you value quickly. At this stage, is it worth it? Well, maybe if you want to waste a few solid hundred hours in trying to get the perfect prompts as well as the credits to test oyur theories.
If you’re experimenting with AI, then be clear on why you need it. If there’s no real alternative that is viable or that you can afford, then consider experimenting with other tools and checking out how to prompt these things well. Like I said, i’ve wasted hundreds of dollars making really garbage prompts that just don’t deliver results.
For my purposes, Canva would have probably been enough, tbh.
AI art has potential, but it’s not quite there yet. Save yourself the headache (and the server costs) for now. Consider exploring other creative options and keep AI in your back pocket for a little while yet..